Configuration for a set of plugins (skills and MCP servers).

Configuration for an MCP server.

Loads and resolves plugin sets from YAML configuration.

Concrete subclass for testing AbstractInstalledAgent.

Configuration for a skill origin.

Handles file diffing operations for test execution.

Root configuration containing all plugin sets.

Installs skills from plugin set configuration.

Generates HTML dashboard for experiment results.

Information about a shared database.

Orchestrator that calls setup functions and configures plugins.

A class for loading and iterating over tasks in a dataset.

Complete set of timeouts for a task execution.

Handles operations related to asciinema recordings, including marker management.

Log formatter for Claude Code agent JSON-lines format.

Parser for Claude agent responses to extract runtime, token usage, and cost metrics.

Parser for Codex agent responses to extract runtime, token usage, and cost metrics.

Exception raised when a debug breakpoint is hit.

Shared utilities for DuckDB data extraction and analysis.

Main converter class that handles the entire conversion pipeline.

Manages file content storage with deduplication using content hashing.

Represents the differences between two file snapshots.

Represents a snapshot of files in a directory at a specific point in time.

Parser for Gemini agent responses to extract runtime, token usage, and cost metrics.

Result from parsing test output, containing both test results and status messages.

Rich-based logger that shows one row per task with live updates.

Configuration for a single solution seed.

Scans an ade-bench tasks directory and returns structured task metadata.

Raised when the LLM response indicates the context length was exceeded.

Manages shared database files and provides copy-on-write functionality.

Base class for agent log formatters.

Parser for Macro agent responses to extract token usage and cost metrics.

Raised when the LLM response cannot be parsed into the expected format.

Structured metadata for an ade-bench task, loaded from task.yaml.

Metadata for a task.

Test dumping tables from DuckDB.

Manages timeout resolution with task-specific overrides. This is the single source of truth for timeout resolution in ADE-Bench. It eliminate

Configuration for a task variant.